# Saved responses

Replies that must go out word for word, with variables filled in and a version per language.

Source: https://docs.omazy.ai/how-to/agent/saved-responses/

Most answers should be written fresh by the agent. Some must not.

A refund policy, a legal line, a regulated disclosure, the exact wording your
compliance team signed off: these need to arrive as written, not as the model's
summary of what was written. That is what saved responses are for. You author
them under **Agent, Saved Responses**.

## What a saved response is

A named block of text that gets sent as-is, with placeholders filled in from the
conversation.

```
Hi {{customer_name}}, your booking {{booking_ref}} is confirmed.
A deposit of {{deposit_amount}} is due within 48 hours.
```

The variables are replaced at send time from what the platform already knows
about the customer and the conversation. If a variable has no value, that is
visible rather than silently blank, because a message reading "Hi ," is worse
than no message.

## Two ways they go out

**The agent uses one** when the situation matches. The brief is where you say
which situations those are.

**An operator sends one** from the inbox. This is the common path for anything
sensitive: a human decides it applies, the wording is still exact.

The second one is worth designing for. A saved response is not only an
automation, it is a way to stop six people writing six slightly different refund
policies at two in the morning.

## Languages

A response holds a version per language, under one name. The agent picks the
version matching the conversation.

This is why deleting a response warns you: it removes every language at once, and
the translation nobody has a copy of is usually the one that took longest.

## When to use one, and when not to

| Use a saved response | Use the brief or knowledge |
|---|---|
| The wording is legally or commercially fixed | The answer should adapt to the question |
| It must be identical every time | It should sound like a conversation |
| Compliance has approved this exact text | You want it to cite a source |

The failure mode is over-using them. An agent answering entirely in saved
responses is a phone tree with better typography, and customers can tell within
two replies. Reserve them for the answers where being exactly right matters more
than being responsive.

## Keeping them honest

Saved responses go stale in the worst way, because they are confident and
specific. A deposit amount hard-coded into the text will be wrong eventually, and
it will be wrong in your own approved wording.

Two habits:

- Put anything numeric in a variable if the platform knows it. Hard-code only
  what genuinely never changes.
- Review them when the underlying policy changes, not on a schedule. The trigger
  is the policy edit, not the calendar.